Folding Forum

Hi everybody,
Why current studies does include any really large proteins like by CPU's?

A protein with twice the number of atoms can run as much as four times longer (but often is somewhat less). FAH started ten years ago with proteins that could be run in a year or so on the cpus of that day and have moved to larger proteins that take a year or so on today's cpus. There's still plenty of room for cpus to get faster or software to work better with multiple processors and bigger proteins are inevitable, but they're something that happens gradually.

(By a year or so, I mean something like 100 Gens of a protein that takes 3 days per Gen or 40 Gens of a protein that runs 10 days. If the scientists know how many Gens a protein is going to need, they don't tell us, and they may not know.)

Thanks codysluder.
I thought, if a medium cpu like my Athlon xp 2400+ can solve a p2611 WU with round 127,000 atoms in ca. 4 days, then a GPU with 38 stream processors and a (theoretically) better performance like my GeF 9500GT, would solve the same WU in less than 4 days.
I know, this is typically linear thinking - usual by us humans - but this is just a simple question..

I think you missed part of what codysluder was saying. Your xp2400+ does not "solve" 127000 atoms in 4 days, it only solves one generation at a time. Each Gen is a small slice of the time it takes for the protein to fold. To "solve" a protein requires many, many sequential generations within each trajectory.

Thanks bruce,
I know what Runs, Clones and Genes mean.
As I wrote to codysluder, my Athlon XP solve (or for better understanding: executes) a WU of that protein in ca. 4 days.
My question is the same: why current studies for GPU does not include(yet) really large proteins like p2611, if some GPU's performs faster?
Once again: I asked "why", not "when"..

You would have to ask the researchers running the projects on the GPUs. It could be the software doesn't work well yet with larger proteins. It could be a hardware limitation. GPUs are very different from CPUs. Or, it could just be that the researches are looking as smaller proteins because their field of study is made up of small proteins.

I couldn't even begin to guess, one or all three, and maybe other reasons... Sorry.

The core for the GPU and for the PS3 are restricted to implicit solvent problems. The classic and SMP clients can run either implicit or explicit solvents. (Why? I don't know, but see the previous post for some possibilities.)

My knowledge of physical chemistry is limited, but if you study the SAME protein, an implicit model is smaller, faster, and less accurate. (or, an explicit model is larger, slower, and more accurate.) Those facts may contribute to a better understanding of why the researchers have made the choices that they've made, but it's still going to be an incomplete answer unless somebody from the Pande Group responds.

I'm going to move this discussion from the GPU forum to the Science forum where they're more likely to respond.

Thanks to bruce and 7im,